Understand Rare Disease by Aggregating In-House Data with Public Data from Pubmed, Social Media, and Search Engine (DSI-SRP)
This DSI-SRP fellowship funded Zheyu (Richard) Zhu to work in the laboratory of Professor Yuankai Huo in the Department of Computer Science during the summer of 2021. Zheyu is a senior with majors in Computer Science and Mathematics.
The project funded by this fellowship aimed to understand the segmentation performance of four different feature extraction methods on a weakly supervised model. Computational pathology, a rapidly growing field, has been proved to be very efficient in therapeutic response prediction and classification of biological subtypes. However, such a method requires a high amount of clinical annotations as a ground truth on whole slide image (WSI) for further processing, which could be extremely time consuming. Here we are using a deep-learning based, weakly supervised pipeline called CLAM, Clustering-constrained Attention Multiple instance learning, to check the validation and accuracy between generated attention section and human-labeled annotation on WSI. To do that, Zheyu and Professor Huo start (1) training CLAM with 937 kidney WSIs in an open-source dataset called The Cancer Genome Atlas (TCGA), which includes three different subtypes, (2) testing the model on 393 kidney WSIs from Pathology AI Platform (PAIP), which includes human annotations, (3) check the result by segmentation accuracy. We evaluated the performance of CLAM by extracting features using different models and batch numbers, such as the ResNet50 model pre-trained by ImageNet and Bit model. As a result, the designed CLAM model reaches an optimal performance of 93.4% on those open-source datasets. The results were exciting and Zheyu and Professor Huo are about to complete a conference paper, which they plan to submit to Society of Photo-Optical Instrumentation Engineers (SPIE), a top journal in the field of renal pathology, in summer 2021.
In addition to receiving support through a DSI-SRP fellowship, this project was supported and facilitated by the DSI Data Science Team through their regular summer workshops and demo sessions.